改进U-Net模型的城市水体精细提取——以洞庭湖为例

Fine Extraction of Urban Water Bodies with Improved U-Net Modeling:Taking Dongting Lake for an Example

针对在大范围、高分辨率遥感影像中传统水体提取方法效率低、微小水体提取效果差的问题,提出一种改进U-Net模型的遥感影像水体精细提取方法。实验以航空高分辨率可见光影像为数据源。分析结果表明,改进的U-Net模型在各项评价指标上均高于经典U-Net模型、基于光谱特征法与基于分类器法,并且改进的网络水体提取结果更加完整,对小目标水体能够准确提取。该研究提高了水体提取语义分割算法的性能,使遥感水体提取工作更加自动化和智能化,不仅验证了基于航空亚米级光学影像在城市水体提取方面的可行性,也为今后相关的研究提供了新的探索思路。

Aiming at the problems of low efficiency of traditional methods for water body extraction and poor extraction effect of tiny water bodies in large-range and high-resolution remote sensing images,this paper proposes an improved U-Net model for fine extraction of water bodies in remote sensing images.The experiment uses aerial high-resolution visible images as the data source,and the results show that the improved U-Net model is higher than the classical U-Net model,the spectral feature-based method and the classifier-based method in terms of IoU and precision rate indexes.Meanwhile,the improved network water body extraction results are more complete and it can accurately extract small target water bodies.This model improves the performance of the semantic segmentation algorithm for water body extraction,and makes the remote sensing water body extraction work more automatic and intelligent.This study not only verifies the feasibility of aerial sub-meter optical image-based extraction of urban water bodies,but also provides new exploration ideas for future related research.